Font Size: a A A

Energy Spectrum Data Processing Of GPU Algorithm Research

Posted on:2014-02-22Degree:MasterType:Thesis
Country:ChinaCandidate:D LiuFull Text:PDF
GTID:2248330398481766Subject:Nuclear technology and applications
Abstract/Summary:PDF Full Text Request
Thesis topics is from the National High Technology Research and Development Program (863Program) Resources in the field of environmental technology Issue Precision spectroscopy detection equipment R&D (No:2012AA061803).Modern instruments spectral analysis, from the characteristic peak of parsing the trend all over the spectrum analysis, greatly increase the amount of calculation. How to improve the ability of data processing to reduce computing time became one of the important indexes of spectrum analytical instrument. In order to improve the efficiency of spectrum analytical instrument, usually for streamlining the optimization algorithm. But is a limited amount of lean based on mathematical method, and the actual execution efficiency is not high on the computer. Another method, is improve the computing capability of the computer. At present, a lot of computer adopts dual core or quad core CPU, its data capacity is greatly strengthened. Simultaneously in the system with the GPU used in image processing, it has the characteristics of parallel computing. Nuclear data processing software, but now most still USES serial single thread is given priority to, not to play multiple data processor of computer computing ability.Due to society’s need for calculation of force and the continuous development of electronic technology, today’s electronic computing system, in addition to the traditional computing devices CPU and other devices, such as the GPU, DSP also gradually into the general category. At present, the high-end single chip GPU single-precision floating-point processing capacity reached1tflops/s has more than ten times the main CPU. Because compared with CPU, GPU, has its unique data parallel computing ability, the GPU has played more and more important role in the field of scientific computing. But different manufacturers of the differences between different types of equipment to develop a lot of difficulty, thus some GPGPU development environment, such as OpenCL, CDUA and DirectCompute, etc. Compared with other two OpenCL support equipment is the most widespread. OpenCL (full Open Computing Language, Open Computing Language) is a heterogeneous system oriented general purpose parallel programming free Open standard, it defines a set of standard application interface API can be used in different equipment. At present common such as Intel multi-core CPU and NVIDIA graphics CARDS, AMD multi-core CPU and graphics can be support.Maximum likelihood algorithm in nuclear detection ability of full spectrum analytical spectral line, can effectively decompose the energy difference is2/3FWHM of overlapping peaks. But due to the large amount of calculation, time-consuming and more restricted the algorithm application. In this paper, the author is to use OpenCL platform, in a heterogeneous system using GPU parallel computing ability, the realization of parallel computing of the maximum likelihood algorithm. In particular platform through test, the algorithm can effectively improve the operation efficiency, the running time for the original1/30, improve the practical feasibility of the algorithm.This thesis mainly research content as follows:1. The research on the energy spectrum analysis the maximum likelihood algorithm application and OpenCL platform.2. Analyses the characteristics of GPU program execution and the feasibility of the maximum likelihood algorithm is run on the GPU, the OpenCL platform mode of GPU programming.3. Complete the GPU to the maximum likelihood calculate spectral line parsing algorithm.4. Test the speed of the program on energy spectrum analysis.
Keywords/Search Tags:Maximum Likelihood method, OpenCL, Parallel Computing
PDF Full Text Request
Related items